Novel model-based decision support system for reliable human machine systems in complex situations

Abstract

Human and machine contribute to the safety of human-machine systems, the interaction between the two is essential for the overall systems safety and reliability. Assuming (formalized) knowledge about the structure of the interactions, monitoring of the human operator becomes possible so that the overall system of human and supervision system can contribute to the safety and reliability of the overall system. The automated recognition of critical situations or the automated detection of human errors allows to intervene in the interaction: to take over the guidance of systems by automation or to warn the human to influence the interaction of human operator and system. Detecting unlogical or errorneous action sequences will help to develop assistance systems to improve the performance of the human-machine system. In the FernBin project beside others the supervision of the captain’s actions in inland shipping is addressed. The focus is to detect human errors and also non-optimal behaviors. Additionally a reliability-based analysis of the captain’s actions will also enable safer driving behaviors and the reduction of accidents and dangerous situations due to suitable warning. In this contribution a Situation-Operator-Modeling (SOM) approach is used to describe the captain-vessel-interaction and to illustrate the captain’s behavior as a graph-based-model. As example a ‘Crossing maneuver’ is considered. A SOM-based action space consisting of possible captain’s behaviors leading to a meaningful desired final situation is developed and applied to a ‘Crossing maneuver’. Using this approach a manifold of sequences can be generated describing the human interaction options to solve the task. Subsequently a modified CREAM approach (cognitive reliability and error analysis method) is used allowing the online calculation of the human reliability performance scores (HPRS), resulting to static reliability measures. The HPRS can be also assigned directly to the SOM-action space, as virtual network of upcoming human actions related to goals. In difference to previous developments this allows beside the deterministic supervision (SOM) also the individualized definition of the safest/most reliable action sequence as well as the opposite. Therefore an event-discretized behavior model situated supervision performance of the captain’s driving behavior with human reliability score numbers is established for the first time. As example the behavior of remotely operating captains of inland vessels is used as experimental example. In this contribution for the first time the newly developed dynamic reliability estimation measure HPRS is combined with the known SOM approach for the captain-vessel interaction example. This example serves as a proof-of-concept example for the new type of assistance system allowing to improve the reliability of the overall system.